基于MFCC和随机森林的GIS动作声纹特征辨识和操作机构异常分类

针对气体绝缘金属封闭开关（GIS）设备的操作机构异常或故障而导致其开关动作时出现分合闸失败或不到位的问题，提出了一种基于梅尔频率倒谱系数（MFCC）和随机森林的GIS设备操作机构异常分类模型。首先，对采集到的声纹信号进行预处理，使用MFCC提取声纹信号的特征；然后，构建随机森林对提取的特征信息进行辨识，得到GIS动作异常的分类结果；最后，以某110 kV的GIS设备为例，采集断路器、隔离开关的储能机构和传动机构异常或故障时的声纹信号，构建了音频样本库，并对所提分类模型与多种经典模型进行了对比测试。结果表明，MFCC能够有效提取出不同工况下GIS动作的声纹信号特征，且随机森林在众多分类识别模型中表现最优，有效提高了GIS动作异常工况识别的准确率。

GIS Action Voiceprint Feature Identification and Operation Mechanism Anomaly Classification Based on MFCC and Random Forest

Aiming at the problem of abnormal or faulty operation mechanism of gas insulated switchgear (GIS), which leads to faults or inability to trip when operating its switches, an abnormal classification model of the operation mechanism of GIS equipment based on the Mel-frequency cepstrum coefficient (MFCC) and random forest is proposed. Firstly, according to the preprocessing of the collected voiceprint signal, MFCC is used to extract the features of the voiceprint signal. Then, a random forest is constructed to identify the voiceprint feature, and the classification results of GIS action anomalies are obtained. Finally, taking a 110 kV GIS equipment as an example, the voiceprint signals of the energy storage mechanism and transmission mechanism of the circuit breaker and the isolating switch are collected when they are abnormal or faulty, and the audio sample library is constructed. The classification model proposed in this paper is compared with a variety of classical models. The results show that MFCC can effectively extract the features of voiceprint signals under different working conditions of GIS actions, and random forest performs best among many classification and recognition models, which can effectively improve the accuracy of abnormal working conditions recognition of GIS actions.